Important Mathematics Institutions | American Institute of Mathematics (AIM)

The simple answer: what mathematical job does AIM perform?

AIM is a problem-centred collaboration engine.

Many conferences are organised around subjects. AIM workshops are deliberately organised around sharper mathematical goals.

The Institute’s own guidance says workshops may aim to make progress on an important unsolved problem, understand the proof of a major new result or examine the convergence of two fields.

This changes the programme design.

Most participants do not give a formal talk. A typical day has only a small number of lectures, followed by discussion sessions and working groups.

The lecture exists to prepare the collaboration; it is not the final product of the meeting.

1994: two mathematics supporters create an independent institute

AIM was established in 1994 by businessmen and Mathematics enthusiasts John Fry and Steve Sorenson.

The founding model was unusual because the institute was not born as a university department or government laboratory.

Its mission was collaboration: advance mathematical knowledge, broaden participation, increase public awareness of mathematical contributions and preserve mathematical history through books and documents.

In 2002, AIM joined the National Science Foundation’s Mathematical Sciences Institutes programme, adding sustained public research support to its philanthropic origins.

Official history: AIM History.

The 2023 move to Caltech changed location without ending independence

After decades in northern California, AIM moved to Pasadena in 2023 and is now housed at Caltech.

The move is institutionally interesting because AIM remained an independent organisation while gaining adjacency to a major research university.

This creates two advantages:

• the institute can preserve its distinctive collaboration model;
• visitors can interact with Caltech mathematicians, physicists, computer scientists and other researchers.

Independence protects programme identity. Proximity expands the network.

Sergei Gukov leads AIM in 2026

The current AIM contact page lists Sergei Gukov as Executive Director.

Gukov is a mathematical physicist known for work linking topology, knot theory, representation theory, quantum field theory and string theory.

His own research history fits AIM’s culture: influential Mathematics can appear when researchers working in different languages are placed around the same structures.

The current leadership also includes Michelle Manes, David Farmer, Brian Conrey and Leslie Hogben, providing depth in arithmetic dynamics, number theory, research programming and collaborative-network design.

Official current contacts: AIM Contact Information.

The AIM workshop is intentionally not a standard conference

AIM workshop proposals are expected to define a specific mathematical focus and explain why the focused format is appropriate.

A standard AIM workshop lasts five days and can support roughly 25–30 researchers.

The schedule deliberately contains relatively few talks.

Afternoons often contain working groups or structured discussion.

This forces organisers to answer a different question from conference organisers:

What can these particular people actually work on together this week?

Official guide: AIM Workshop Proposal Guide.

Open problems are collected before participants arrive

For AIM-style workshops, participants may be asked to suggest open problems and research questions before the meeting begins.

This is a small procedural detail with large consequences.

The workshop begins with an explicit shared problem landscape rather than waiting for useful questions to emerge accidentally after several days.

The problem list can contain both:

• specific questions where progress during the week is plausible;
• larger questions that may organise research long after the workshop ends.

The institution therefore makes unfinished Mathematics visible before asking researchers to collaborate.

28 September 2026: AI agents become the next frontier

The next AIM workshop, running from 28 September to 2 October 2026, is Mathematical Foundations for AI Agents in Complex Environments.

The programme begins from a limitation in current AI development.

Many systems are optimised on relatively static benchmarks. Real deployed agents operate inside environments that react.

Other agents change behaviour. Users adapt. Incentives shift. Data distributions evolve. Actions change future observations.

This creates mathematical questions involving:

• dynamical systems;
• game theory;
• control theory;
• online learning;
• stochastic processes;
• multi-agent interaction;
• robust optimisation;
• generalisation under non-stationarity.

Official current workshop: Mathematical Foundations for AI Agents in Complex Environments.

Static benchmarks can conceal dynamic failure

A predictive system can look excellent when the environment does not respond to it.

But an agent changes the environment through action.

A recommender changes what users watch. A trading agent changes prices. A routing algorithm changes congestion. A tutoring system changes what a learner knows.

The mathematical object is therefore a feedback system rather than a one-shot predictor.

state → observation → action → changed state → new observation → next action.

Once the loop is recognised, control, games and dynamics become as important as prediction accuracy.

AIM had already connected AI to number theory earlier in 2026

In May 2026, AIM held a workshop on AI and Number Theory.

The organisers asked two broad questions:

• Which research problems in Mathematics are amenable to AI?
• How can AI itself be improved so that it becomes more useful for serious mathematical research?

The workshop focused on improving theorem bounds and algorithms in number theory and aimed to create benchmark problems representing the actual limits of current AI systems.

This is a sophisticated move.

Generic benchmark success can hide weaknesses. Research mathematicians can create tasks whose difficulty reflects real mathematical structure rather than pattern imitation.

Official workshop: AI and Number Theory.

9 October 2026: Terry Tao will lecture on machine-assisted proof

AIM’s 2026 public lecture is scheduled for 9 October 2026, when Terry Tao will speak on Machine Assisted Proof.

The title places AIM directly inside one of the most important emerging transitions in Mathematics.

Machines can assist proof through several mechanisms:

• formal proof assistants;
• symbolic computation;
• computer algebra;
• certified numerical bounds;
• large finite verification;
• AI-assisted lemma search and formalisation.

The frontier question is not whether a computer touched the proof.

The frontier question is whether the final mathematical dependency chain is trustworthy.

SQuaREs create a second collaboration scale

AIM’s SQuaRE programme supports groups of roughly four to six mathematicians pursuing an ambitious research project.

A group spends a week at AIM and may return in later years, with support available for up to three meetings over three consecutive years.

This solves a different problem from a workshop.

A workshop may discover the collaboration. A SQuaRE can sustain it.

Workshop → find the people and problem.
SQuaRE → return until the problem becomes a paper.

Official programme: AIM SQuaREs.

2026 SQuaREs show the breadth of small-team Mathematics

Current and recent 2026 SQuaRE activity includes research in:

• quaternionic modular forms;
• symmetric tensor categories in positive characteristic;
• sectional curvature of solvmanifolds;
• algebraic geometry of chemical reaction networks;
• wave turbulence and dispersion.

The range is important because the SQuaRE format is method-neutral.

The institution supplies concentrated collaboration. The mathematical community decides which frontier needs it.

AIM Research Communities create a third, virtual timescale

AIM Research Communities, or ARCs, involve at least forty participants and operate primarily online.

They can take several forms:

• research groups producing mathematical results;
• learning communities for graduate students and junior researchers;
• networks designed to build a field and connect researchers with different backgrounds.

Many persist for years.

The virtual format solves a structural problem that physical workshops cannot.

A researcher may have no nearby collaborator in their speciality and may not be able to travel repeatedly. A long-lived online community can keep the field socially connected between physical meetings.

Official programme: AIM Research Communities.

Current ARCs reveal where distributed collaboration is useful

Current AIM Research Communities include programmes in quantum research, graph polynomials and invariants, digital ecosystems for Mathematics, infectious disease and ecology, climate Mathematics, Fourier restriction, number theory, inverse eigenvalue problems and representation theory.

This breadth reinforces the layered AIM architecture.

A week-long workshop is high intensity. A SQuaRE is a small-team recurring collaboration. An ARC is lower intensity but can persist for years.

Different mathematical problems require different collaboration frequencies.

The workshop format is designed around working groups

AIM explicitly distinguishes its workshop model from Oberwolfach.

Both use small groups and week-long concentration, but AIM generally uses fewer talks and more organised discussion sessions and collaborative work.

The distinction is architectural rather than hierarchical.

Oberwolfach optimises an intensive specialist meeting culture. AIM makes the explicit open problem and the working group especially central.

AIM keeps an open application route into workshops

AIM workshops reserve places for applicants rather than filling every seat exclusively by invitation.

Successful applicants receive support for travel and accommodation.

This matters because a frontier community can become closed if participation depends only on already knowing the organisers.

An application route creates a mechanism by which a researcher outside the immediate social network can demonstrate relevant expertise and enter the collaboration.

Mathematical foundations of AI agents is a systems problem

The upcoming AI-agents workshop also demonstrates why the frontier increasingly connects Mathematics to systems thinking.

An intelligent agent embedded in a complex environment may need to reason about:

• partial observability;
• other strategic actors;
• uncertain dynamics;
• long horizons;
• changing objectives;
• safety constraints;
• distribution shifts.

This is not one optimisation problem. It is a nested control-and-inference problem.

The workshop therefore needs researchers from learning theory, optimisation, control, probability and game theory to share a language.

AIM can turn mathematics education into research infrastructure too

AIM also supports Math Circles, Research Experiences for Undergraduate Faculty and programmes designed to improve the research pipeline.

The 2026 REUF workshop, for example, was designed to help faculty at undergraduate institutions involve their students in active research and to renew faculty research collaborations.

This is an important extension of the frontier model.

A country’s research system is weakened if advanced collaboration exists only at universities already rich in research resources.

Research infrastructure can be designed to propagate outward.

Mathematical history is part of AIM’s mission

AIM’s mission also includes preserving the history of Mathematics through books and documents.

This may appear separate from collaborative workshops, but the two activities share a principle: Mathematics is cumulative.

Open problems depend on earlier definitions. New proofs often rediscover forgotten techniques. Historical documents show how ideas actually developed rather than how textbooks later simplified the story.

A frontier institution therefore benefits from remembering the old frontier accurately.

What a Secondary or JC student can learn from AIM

1. Mathematics is not mostly listening

AIM deliberately limits lectures because research requires working on problems, not only hearing about them.

2. A problem list can organise learning

Researchers often learn a field because an open problem tells them which tools they need next.

3. Small groups can outperform large crowds for technical work

SQuaREs exist because four or five committed collaborators can sometimes make more progress than a hundred-person meeting.

4. AI does not remove the need for proof

AI-assisted exploration and machine-assisted proof increase the need for precise verification, not decrease it.

5. Collaboration can be designed

The number of talks, size of group, application process and working sessions all change what mathematical interactions are possible.

From school Mathematics toward AIM frontiers

• Functions and iteration → dynamical systems → AI agents and control.
• Probability → stochastic processes → learning in uncertain environments.
• Graphs → graph invariants → combinatorics and network Mathematics.
• Number theory → L-functions and arithmetic algorithms → AI-assisted research benchmarks.
• Geometry → curvature and topology → modern geometric workshops.
• Proof → formal logic and computation → machine-assisted proof.

AIM institutional map

Entity	American Institute of Mathematics (AIM)
Type	Independent nonprofit mathematical sciences research institute; NSF Mathematical Sciences Institute
Founded	1994
Founders	John Fry and Steve Sorenson
Current location	Richard N. Merkin Center, Caltech campus, Pasadena, California
Moved to Pasadena	2023
Current Executive Director checked	Sergei Gukov
Current Deputy Director checked	Michelle Manes
Core research mechanisms	Focused workshops; SQuaREs; AIM Research Communities
Annual physical participation	Approximately 1,000 mathematicians in workshops and SQuaREs
Next workshop	Mathematical Foundations for AI Agents in Complex Environments, 28 September–2 October 2026
Upcoming public lecture	Terry Tao, Machine Assisted Proof, 9 October 2026
Verification date	25 September 2026

Connections into the Bukit Timah Tutor Mathematics estate

• Automorphic Representations, Harmonic Analysis, L-Functions and the Langlands Program
• Prime Sieves, Probable Primes and Primality Testing
• Representation Theory in Mathematics
• Manifolds, Knots and Geometric Topology
• Construction, Verification, Witnesses and Impossibility
• Quantum Complexity Theory

Return to the Singapore Mathematics Hub.

Official AIM sources

• American Institute of Mathematics — Homepage
• Mission and History
• Current Staff and Contact
• AIM Workshops
• Workshop Proposal Guide
• SQuaREs
• AIM Research Communities
• Mathematical Foundations for AI Agents in Complex Environments
• AI and Number Theory

The larger lesson

The American Institute of Mathematics demonstrates that collaboration itself can be engineered as a mathematical resource.

The Institute does not assume that putting experts in the same room is enough.

It limits the group. It limits the talks. It exposes open problems early. It creates working sessions. It leaves application routes open. It gives successful collaborations a smaller SQuaRE format in which to continue. It gives distributed fields a virtual Research Community in which to remain connected for years.

That architecture is especially relevant as AI enters research Mathematics. Faster conjecture generation, faster coding and machine-assisted proof will increase the number of candidate ideas. The scarce resource may increasingly become expert attention capable of deciding which ideas are mathematically meaningful.

AIM matters because it organises mathematicians around the unfinished problem rather than around the performance of already-finished Mathematics.

That makes the American Institute of Mathematics an essential node in any serious map of modern frontier Mathematics.